Delivery Forecasting

Predict delivery dates for in-progress features using historical throughput, with P50 and P85 scenario projections and at-risk PR detection.

Overview

Delivery Forecasting predicts when in-progress work will ship based on your team's actual historical throughput — not velocity averages or story point estimates. It ingests the last 16 weeks of weekly PR merge data and runs a Monte Carlo simulation to produce a range of delivery scenarios, giving you honest confidence intervals instead of a single optimistic date.

Forecasting is available to all roles. It is team-scoped: each team sees its own projection based on its own throughput history.

How Forecasting Works

Koalr samples your team's weekly merged PR count over the last 16 weeks and treats that distribution as the input to a Monte Carlo simulation. Each simulation run draws a random weekly throughput from the historical distribution and advances the backlog until it clears. After 10,000 simulation runs, the output is a full probability distribution of completion dates.

Two scenarios are surfaced:

Scenario	Percentile	Interpretation
P50	50th percentile	Equally likely to finish before or after this date. Use for internal planning.
P85	85th percentile	85% of simulations finish by this date. Use for external commitments.

The simulation respects your team's observed variance — a team with highly consistent throughput will produce tight date ranges; a team with volatile output will produce wide ones.

What-If Scenarios

Use the interactive sliders on the Forecasting page to model hypothetical changes before committing to them:

Slider	Effect
Add capacity	Increases simulated weekly throughput proportionally
Remove capacity	Decreases simulated weekly throughput proportionally
Reduce scope	Removes a fixed number of backlog items from the simulation
Add scope	Adds items to the backlog and reruns the simulation

Each adjustment immediately rerenders the P50 and P85 date cards. No changes are saved unless you explicitly export or share the scenario.

At-Risk PR Detection

A PR is flagged as at-risk when it has been open for more than 2x the team's P75 cycle time. The at-risk list appears below the forecast chart and is sorted by age descending.

Each row shows the PR title, author, current age, and how far over the threshold it is. Clicking a row opens the PR detail drawer with review timeline and review activity.

Resolving at-risk PRs directly improves forecast accuracy by keeping throughput data clean and reducing carry-over drag between weeks.

Forecast Accuracy

After 30 or more days of use, the Forecasting page shows a historical accuracy panel that overlays past predictions against actual ship dates. This lets you calibrate how much buffer your team typically needs and whether the P50 or P85 scenario is more reliable for your planning style.

Accuracy data updates once per week.

Deadline Mode

If you have an external deadline, switch to Deadline Mode by entering a target date in the date picker. Instead of projecting a completion date, the simulation runs in reverse and reports:

How many backlog items fit within the window at P50
How many fit at P85
The implied scope cut needed to hit the deadline with 85% confidence

Deadline Mode is useful for sprint planning reviews and quarterly commitment conversations.

Limitations

Forecasting works best with 8 or more weeks of merged PR data. With fewer than 8 weeks, the simulation uses the available data but displays a low-confidence warning.
Teams with fewer than 3 active developers often show high variance in weekly throughput. The resulting date ranges may be too wide to be actionable until more history accumulates.
Forecasting is based on PR throughput, not story points or issue estimates. If your team merges many small PRs per feature, forecasts will be more accurate than teams that batch large PRs.
The simulation does not account for planned holidays or scheduled leave. Adjust scope sliders manually to compensate.

DORA Metrics
Work Log
Teams and Members